[PATCH, AARCH64] Optimized memset
Fri Jul 17 17:46:00 GMT 2015
> Marcus Shawcroft wrote:
> On 10 July 2015 at 13:44, Wilco Dijkstra <email@example.com> wrote:
> > It turns out to be slightly faster on A53 too, so I've updated my patch.
> > Wilco
> > ChangeLog:
> > 2015-07-10 Wilco Dijkstra <firstname.lastname@example.org>
> > * newlib/libc/machine/aarch64/memset.S (memset):
> > Rewrite of optimized memset.
> > OK for commit?
> Wilco, This patch appears to cause various aarch64_be-none-elf gcc
> testsuite tests to drop into infinite loops. One example is:
> RUNTESTFLAGS="builtins.exp=fprintf.c" make check-gcc
> would you mind investigating please?
> I'm going to revert the patch while we investigate this.
It turns out there was a typo in this instruction due to testing for
different zva values:
and tmp1w, tmp1w, 4 (should be 15)
As a result memset fails if you use a dz zva value other than the
common setting of 64 bytes. With the change it correctly works with
any legal setting, and aarch64_be-none-elf passes.
Attached is the fixed version of the patch.
-------------- next part --------------
An embedded and charset-unspecified text was scrubbed...
More information about the Newlib